HumanST-46M
收藏资源简介:
HumanST-46M 是用于空间转录组学层次基础模型 NexuST 的预训练数据集。该数据集包含 76 个 H5AD 格式的文件,其中训练集 72 个(约 94.84 GiB),验证集 4 个(约 4.79 GiB),总大小约 99.63 GiB。数据保留原始平台子目录和文件名,覆盖 Xenium、MERFISH 和 CosMx 等平台生成的样本。验证集包括肺纤维化 Xenium 数据、成人脑 MERFISH 数据以及正常和肝癌 CosMx 肝脏数据。数据集仅提供 H5AD 文件,不包含 CellWorld 内存映射表示、采样索引或下游任务数据。用户可通过 Hugging Face Hub 使用 snapshot_download 下载指定子集。该数据集适用于空间转录组学的模型预训练、表示学习及相关生物信息学研究。
HumanST-46M is a pretraining dataset for the hierarchical foundation model NexuST for spatial transcriptomics. It contains 76 H5AD files, including 72 training files (approximately 94.84 GiB) and 4 validation files (approximately 4.79 GiB), with a total size of about 99.63 GiB. The data retains original platform subdirectories and filenames, covering samples generated from platforms such as Xenium, MERFISH, and CosMx. The validation set includes Xenium data of lung fibrosis, MERFISH data of adult brain, and CosMx liver data of normal and hepatocellular carcinoma. The dataset only provides H5AD files and does not include CellWorld memory-mapped representations, sampling indices, or downstream task data. Users can download specified subsets via snapshot_download from the Hugging Face Hub. This dataset is suitable for model pretraining, representation learning, and related bioinformatics research in spatial transcriptomics.
HumanST-46M 数据集概述
基本信息
- 数据集名称:HumanST-46M
- 数据页面:https://huggingface.co/datasets/Haiping-UoM/HumanST-46M
- 语言:英语(en)
- 标签:spatial-transcriptomics、biology、h5ad、anndata
- 论文:NexuST: A Hierarchical Foundation Model for Spatial Transcriptomics(bioRxiv, 2026),DOI:https://doi.org/10.64898/2026.09.22.753590
- 用途:提供 NexuST 项目所使用的预训练 H5AD 数据
数据集构成
本发布版共包含 76 个 H5AD 文件,训练数据与验证数据分别存放于不同目录。
| 划分 | H5AD 文件数 | 字节数 | 大小(GiB) |
|---|---|---|---|
train/ |
72 | 101,834,973,232 | 94.84 |
val/ |
4 | 5,138,760,318 | 4.79 |
| 合计 | 76 | 106,973,733,550 | 99.63 |
文件保留原有的平台子目录与文件名。本发布版仅包含 H5AD 数据;CellWorld 内存映射表示、采样索引以及单独的下游数据集均不包含在内。
验证集文件
val/xenium/Pulmonary_Fibrosis/pulmonary_fibrosis_xenium.h5adval/merfish/adult_umb5958.h5adval/cosmx/cosmx_liver_normal.h5adval/cosmx/cosmx_liver_cancer.h5ad
下载方式
python from huggingface_hub import snapshot_download
将 train/ 和 val/ 下载到 HumanST-46M/h5ad/,以匹配本地项目结构。
snapshot_download( repo_id="Haiping-UoM/HumanST-46M", repo_type="dataset", allow_patterns=["train//*.h5ad", "val//*.h5ad"], local_dir="HumanST-46M/h5ad", )
若仅下载验证集,可使用 allow_patterns=["val/**/*.h5ad"]。
读取单个文件
使用 AnnData 读取单个文件:
python import anndata as ad
adata = ad.read_h5ad( "HumanST-46M/h5ad/val/cosmx/cosmx_liver_normal.h5ad", backed="r", ) print(adata.shape) adata.file.close()
相关仓库
- NexuST 模型:https://huggingface.co/Haiping-UoM/NexuST
- NexuST 模型代码:https://github.com/UoM-HealthAI/NexuST
文档状态
- 上述文件数量与大小均基于本发布版测量。
- 数据集名称沿用自项目;此处尚未记录精确的细胞计数审计。
- 来源归属、预处理、字段描述及许可条款将在核实后单独记录。
引用
bibtex @article{liu2026nexust, title = {NexuST: A Hierarchical Foundation Model for Spatial Transcriptomics}, author = {Liu, Haiping and Zhao, Qian and Lin, Lijing and Zou, Zhiyong and Cai, Wenhao and Sun, Jingyuan and Zhou, Yuxi and Alvarez, Mauricio A. and Gilmore, Andrew and Rattray, Magnus and Frangi, Alejandro F. and Zhou, Hongpeng}, journal = {bioRxiv}, year = {2026}, doi = {10.64898/2026.09.22.753590} }





